Line data Source code
1 : /**
2 : * Copyright (c) 2025 Huawei Technologies Co., Ltd.
3 : * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
4 : * CANN Open Software License Agreement Version 2.0 (the "License").
5 : * Please refer to the License for details. You may not use this file except in compliance with the License.
6 : * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
7 : * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
8 : * See LICENSE in the root of the software repository for the full text of the License.
9 : */
10 :
11 : /*!
12 : * \file kernel_check_vec_reduce_util.h
13 : * \brief
14 : */
15 :
16 : #ifndef ASCENDC_CHECK_VEC_REDUCE_UTIL_H
17 : #define ASCENDC_CHECK_VEC_REDUCE_UTIL_H
18 : #if ASCENDC_CPU_DEBUG
19 : #include <string>
20 : #include "kernel_utils.h"
21 : namespace AscendC {
22 : namespace check {
23 : struct VecReduceApiParams {
24 : VecReduceApiParams() {}
25 72 : VecReduceApiParams(
26 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint32_t dstDtypeBytesIn, uint32_t src0DtypeBytesIn, int32_t repeatIn,
27 : uint16_t dstRepStrideIn, uint16_t srcBlkStrideIn, uint16_t srcRepStrideIn, uint64_t dstSizeIn,
28 : uint64_t src0SizeIn, uint8_t dstPosIn, uint8_t src0PosIn)
29 72 : {
30 72 : dstAddr = dstAddrIn;
31 72 : src0Addr = src0AddrIn;
32 72 : dstDtypeBytes = dstDtypeBytesIn;
33 72 : src0DtypeBytes = src0DtypeBytesIn;
34 72 : repeatTimes = repeatIn;
35 72 : dstRepeatStride = dstRepStrideIn;
36 72 : src0BlockStride = srcBlkStrideIn;
37 72 : src0RepeatStride = srcRepStrideIn;
38 72 : dstSize = dstSizeIn;
39 72 : src0Size = src0SizeIn;
40 72 : dstLogicPos = dstPosIn;
41 72 : src0LogicPos = src0PosIn;
42 72 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
43 72 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
44 72 : }
45 28 : VecReduceApiParams(
46 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint64_t src1AddrIn, uint32_t dstDtypeBytesIn,
47 : uint32_t src0DtypeBytesIn, uint32_t src1DtypeBytesIn, int32_t repeatIn, uint32_t calCountIn, bool calIndexIn,
48 : uint64_t dstSizeIn, uint64_t src0SizeIn, uint64_t src1SizeIn, uint8_t dstPosIn, uint8_t src0PosIn,
49 : uint8_t src1PosIn)
50 28 : {
51 28 : dstAddr = dstAddrIn;
52 28 : src0Addr = src0AddrIn;
53 28 : src1Addr = src1AddrIn;
54 28 : dstDtypeBytes = dstDtypeBytesIn;
55 28 : src0DtypeBytes = src0DtypeBytesIn;
56 28 : src1DtypeBytes = src1DtypeBytesIn;
57 28 : repeatTimes = repeatIn;
58 28 : calCount = calCountIn;
59 28 : calIndex = calIndexIn;
60 28 : dstSize = dstSizeIn;
61 28 : src0Size = src0SizeIn;
62 28 : src1Size = src1SizeIn;
63 28 : dstLogicPos = dstPosIn;
64 28 : src0LogicPos = src0PosIn;
65 28 : src1LogicPos = src1PosIn;
66 28 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
67 28 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
68 28 : src1Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src1PosIn)));
69 28 : }
70 :
71 4 : VecReduceApiParams(
72 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint64_t src1AddrIn, uint32_t dstDtypeBytesIn,
73 : uint32_t src0DtypeBytesIn, uint32_t src1DtypeBytesIn, int32_t repeatIn, uint32_t calCountIn, uint64_t dstSizeIn,
74 : uint64_t src0SizeIn, uint64_t src1SizeIn, uint8_t dstPosIn, uint8_t src0PosIn, uint8_t src1PosIn)
75 4 : {
76 4 : dstAddr = dstAddrIn;
77 4 : src0Addr = src0AddrIn;
78 4 : src1Addr = src1AddrIn;
79 4 : dstDtypeBytes = dstDtypeBytesIn;
80 4 : src0DtypeBytes = src0DtypeBytesIn;
81 4 : src1DtypeBytes = src1DtypeBytesIn;
82 4 : repeatTimes = repeatIn;
83 4 : calCount = calCountIn;
84 4 : dstSize = dstSizeIn;
85 4 : src0Size = src0SizeIn;
86 4 : src1Size = src1SizeIn;
87 4 : dstLogicPos = dstPosIn;
88 4 : src0LogicPos = src0PosIn;
89 4 : src1LogicPos = src1PosIn;
90 4 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
91 4 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
92 4 : src1Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src1PosIn)));
93 4 : }
94 :
95 4 : VecReduceApiParams(
96 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint32_t dstDtypeBytesIn, uint32_t src0DtypeBytesIn,
97 : uint32_t calCountIn, uint64_t dstSizeIn, uint64_t src0SizeIn, uint8_t dstPosIn, uint8_t src0PosIn)
98 4 : {
99 4 : dstAddr = dstAddrIn;
100 4 : src0Addr = src0AddrIn;
101 4 : dstDtypeBytes = dstDtypeBytesIn;
102 4 : src0DtypeBytes = src0DtypeBytesIn;
103 4 : calCount = calCountIn;
104 4 : dstSize = dstSizeIn;
105 4 : src0Size = src0SizeIn;
106 4 : dstLogicPos = dstPosIn;
107 4 : src0LogicPos = src0PosIn;
108 4 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
109 4 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
110 4 : }
111 :
112 0 : VecReduceApiParams(
113 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint64_t src1AddrIn, uint32_t dstDtypeBytesIn,
114 : uint32_t src0DtypeBytesIn, uint32_t src1DtypeBytesIn, int32_t repeatIn, bool calIndexIn, uint64_t dstSizeIn,
115 : uint64_t src0SizeIn, uint64_t src1SizeIn, uint8_t dstPosIn, uint8_t src0PosIn, uint8_t src1PosIn,
116 : uint16_t src0RepeatStrideIn)
117 0 : {
118 0 : dstAddr = dstAddrIn;
119 0 : src0Addr = src0AddrIn;
120 0 : src1Addr = src1AddrIn;
121 0 : dstDtypeBytes = dstDtypeBytesIn;
122 0 : src0DtypeBytes = src0DtypeBytesIn;
123 0 : src1DtypeBytes = src1DtypeBytesIn;
124 0 : repeatTimes = repeatIn;
125 0 : calIndex = calIndexIn;
126 0 : dstSize = dstSizeIn;
127 0 : src0Size = src0SizeIn;
128 0 : src1Size = src1SizeIn;
129 0 : dstLogicPos = dstPosIn;
130 0 : src0LogicPos = src0PosIn;
131 0 : src1LogicPos = src1PosIn;
132 0 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
133 0 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
134 0 : src1Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src1PosIn)));
135 0 : src0RepeatStride = src0RepeatStrideIn;
136 0 : }
137 :
138 8 : VecReduceApiParams(
139 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint64_t src1AddrIn, uint32_t dstDtypeBytesIn,
140 : uint32_t src0DtypeBytesIn, uint32_t src1DtypeBytesIn, int32_t repeatIn, uint64_t dstSizeIn, uint64_t src0SizeIn,
141 : uint64_t src1SizeIn, uint8_t dstPosIn, uint8_t src0PosIn, uint8_t src1PosIn, uint16_t src0RepeatStrideIn)
142 8 : {
143 8 : dstAddr = dstAddrIn;
144 8 : src0Addr = src0AddrIn;
145 8 : src1Addr = src1AddrIn;
146 8 : dstDtypeBytes = dstDtypeBytesIn;
147 8 : src0DtypeBytes = src0DtypeBytesIn;
148 8 : src1DtypeBytes = src1DtypeBytesIn;
149 8 : repeatTimes = repeatIn;
150 8 : dstSize = dstSizeIn;
151 8 : src0Size = src0SizeIn;
152 8 : src1Size = src1SizeIn;
153 8 : dstLogicPos = dstPosIn;
154 8 : src0LogicPos = src0PosIn;
155 8 : src1LogicPos = src1PosIn;
156 8 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
157 8 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
158 8 : src1Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src1PosIn)));
159 8 : src0RepeatStride = src0RepeatStrideIn;
160 8 : }
161 :
162 : uint64_t dstAddr = 0;
163 : uint64_t src0Addr = 0;
164 : uint64_t src1Addr = 0;
165 : uint8_t repeatTimes = 0;
166 : uint16_t dstBlockStride = 0;
167 : uint16_t src0BlockStride = 0;
168 : uint16_t src1BlockStride = 0;
169 : uint16_t dstRepeatStride = 0;
170 : uint16_t src0RepeatStride = 0;
171 : uint16_t src1RepeatStride = 0;
172 : uint32_t dstDtypeBytes = 0;
173 : uint32_t src0DtypeBytes = 0;
174 : uint32_t src1DtypeBytes = 0;
175 : uint64_t dstSize = 0;
176 : uint64_t src0Size = 0;
177 : uint64_t src1Size = 0;
178 : uint8_t dstLogicPos = 0;
179 : uint8_t src0LogicPos = 0;
180 : uint8_t src1LogicPos = 0;
181 : uint8_t dstPos = 0;
182 : uint8_t src0Pos = 0;
183 : uint8_t src1Pos = 0;
184 : uint32_t calCount = 0;
185 : bool calIndex = 0;
186 : };
187 :
188 : struct VecReduceWhlApiParams {
189 : VecReduceWhlApiParams() {}
190 36 : VecReduceWhlApiParams(
191 : uint64_t dstAddrIn, uint64_t src0AddrIn, uint32_t dstDtypeBytesIn, uint32_t src0DtypeBytesIn, int32_t repeatIn,
192 : uint16_t dstRepStrideIn, uint16_t srcBlkStrideIn, uint16_t srcRepStrideIn, ReduceOrder orderIn,
193 : uint64_t dstSizeIn, uint64_t src0SizeIn, uint8_t dstPosIn, uint8_t src0PosIn)
194 36 : {
195 36 : dstAddr = dstAddrIn;
196 36 : src0Addr = src0AddrIn;
197 36 : dstDtypeBytes = dstDtypeBytesIn;
198 36 : src0DtypeBytes = src0DtypeBytesIn;
199 36 : repeatTimes = repeatIn;
200 36 : dstRepeatStride = dstRepStrideIn;
201 36 : src0BlockStride = srcBlkStrideIn;
202 36 : src0RepeatStride = srcRepStrideIn;
203 36 : order = orderIn;
204 36 : dstSize = dstSizeIn;
205 36 : src0Size = src0SizeIn;
206 36 : dstLogicPos = dstPosIn;
207 36 : src0LogicPos = src0PosIn;
208 36 : dstPos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(dstPosIn)));
209 36 : src0Pos = static_cast<uint8_t>(GetPhyType(static_cast<TPosition>(src0PosIn)));
210 36 : }
211 :
212 : uint64_t dstAddr = 0;
213 : uint64_t src0Addr = 0;
214 : uint8_t repeatTimes = 0;
215 : uint16_t src0BlockStride = 0;
216 : uint16_t dstRepeatStride = 0;
217 : uint16_t src0RepeatStride = 0;
218 : uint32_t dstDtypeBytes = 0;
219 : uint32_t src0DtypeBytes = 0;
220 : ReduceOrder order = ReduceOrder::ORDER_VALUE_INDEX;
221 : uint64_t dstSize = 0;
222 : uint64_t src0Size = 0;
223 : uint8_t dstLogicPos = 0;
224 : uint8_t src0LogicPos = 0;
225 : uint8_t dstPos = 0;
226 : uint8_t src0Pos = 0;
227 : };
228 :
229 : bool CheckFunReduceImplForMaskArray(VecReduceApiParams& chkParams, const uint64_t mask[], const char* intriName);
230 : bool CheckFunReduceImpl(VecReduceApiParams& chkParams, const uint64_t mask, const char* intriName);
231 : bool CheckFunReduceImpl(VecReduceApiParams& chkParams, const char* intriName);
232 :
233 : bool CheckFunReduceOtherImplForMaskArray(VecReduceApiParams& chkParams, const uint64_t mask[], const char* intriName);
234 : bool CheckFunReduceOtherImpl(VecReduceApiParams& chkParams, const uint64_t mask, const char* intriName);
235 :
236 : bool CheckFunReduceOtherWhlImplForMaskArray(
237 : VecReduceWhlApiParams& chkParams, const uint64_t mask[], const char* intriName);
238 : bool CheckFunReduceOtherWhlImpl(VecReduceWhlApiParams& chkParams, const uint64_t mask, const char* intriName);
239 : bool CheckFunReduceImplMode2(VecReduceApiParams& chkParams, const char* intriName);
240 :
241 : } // namespace check
242 : } // namespace AscendC
243 : #endif
244 : #endif
|
